maximecb
commented
14 hours ago I agree with everything you said.
Possibly we could just remove the "speed comparison results over time?" Like you said it's not very useful and it's kind of confusing to have multiple graphs that are similar but not the same. Better to have fewer graphs that are more likely to be what people should actually be looking at? Similar feelings about "YJIT Results-Over-Time Deep Dive".
Also, what the hell, total time in seconds to run the benchmarks? Summed across all iterations? But why...???
The timeline links at the bottom (and the pages they link to) could use descriptions.
For example, the "speed comparison results" one is useful for investigating changes but for overall YJIT performance the "yjit speedup" is a much better graph.
The time measured for any given benchmark means very little by itself (and changes when we change hardware or benchmark code). Rather the relative comparison to other benchmarks from the same run is the metric most people would want to watch.